Proceedings of the California Avocado Research Symposium, November 4, 2006. University of California,
Riverside. Sponsored by the California Avocado Commission. Pages 57-63.
Assessing the Genetic Determination of Valuable Avocado Traits Using
Microsatellite (SSR) Markers and Quantitative Trait Locus (QTL) Analysis
Ongoing Project: Year 5 of 8
Project Leader: Michael T. Clegg (949) 824-4490, email: firstname.lastname@example.org
Department of Ecology and Evolutionary Biology, UC Irvine, and
Department of Botany and Plant Sciences, UC Riverside
Vanessa Ashworth (UCI), Haofeng Chen (UCR), Shizhong Xu (UCR)
Varietal improvement in avocado (Persea americana Mill.) has long relied on multi-year field
trials during which large numbers of seedlings are grown to maturity and compared for desirable
characteristics. Inferior trees are removed from the breeding block as their deficiencies become
apparent, leaving only the most promising genotypes. However, the time, land resource, and
labor costs associated with growing trees to the appropriate stage of development are
considerable. The pace of varietal improvement would accelerate substantially through the
application of molecular markers that are detectable using DNA extracted from seedlings. If
transmitted along with desirable traits, the markers can be used as surrogates for these traits and
can be applied quickly to a large number of seedlings to enrich the initial pool of trees for traits
that are of interest to the Industry. Our research is designed to identify markers that are co-
transmitted with genetic factors conferring desirable characteristics in avocado. Our objectives
are (1) to link avocado traits of interest to growers with molecular markers and (2) to harness this
information via marker-assisted selection. This marker-guided method of variety improvement
has the potential to increase selection intensity by several orders of magnitude.
Benefits to the Industry
• By gaining access to a less stochastic, faster, and less land- and labor-intensive method of
breeding avocado, called marker-assisted selection (MAS), and to allied breeding
• Our molecular marker data is a permanent resource: it can be combined with any trait of
interest measured (non-destructively) in our population of experimental trees to identify
marker-trait associations. This will furnish more detailed knowledge of the genetics of
both complex traits (QTLs), as well as mono- and oligogenic traits
• Once the molecular framework is in place, other molecular studies can be more easily
piggy-backed, thus magnifying the returns on initial investment
Unraveling the association between desirable traits and molecular markers relies on (1) the
availability of a pool of molecular markers, and (2) a replicated experimental population of trees
having a known genetic constitution. A summary of marker- and tree-related information is
presented in Table 1.
Table 1. Summary of facts and figures relating to our markers and experimental trees.
1. 205 distinct genotypes of open-pollinated ‘Gwen’ progeny
2. Four clones of each genotype at SCREC (Irvine) and Agricultural Operations (AgOps,
Riverside); two clones planted at each location
3. Grafted onto Duke 7 rootstock
4. Trees planted out in 2001 (SCREC) and 2002 (AgOps)
5. 398 and 285 trees (= 683 trees) at each site, respectively
6. 127 microsatellite markers
7. 364 and 161 trees (525 trees) bore fruit this year
8. 34.7% of genotypes were sired by ‘Bacon’, 39.8% by ‘Fuerte’ and 25.5% by ‘Zutano’
9. Fruit dry weights (March–April): 15.6–43.8%, averaging 29.3 ± 5.0% DW at SCREC and
33.5 ± 5.4% at AgOps
10. Fruit weights: <100 g to 799 g, with an average of 281 g (SCREC) and 255 g (AgOps)
11. Fruit load/tree: high (>100 fruit/tree) in 27.3% of trees, medium (50–99 fruit/tree) in
38.3%, low (1–49 fruit/tree) in 26%, and 8.5% of trees bore no fruit.
Key to this project is the maintenance of four clones of each experimental tree genotype growing
at two different locations. Replicated field trials are essential because the environment exerts
nongenetic effects on all traits. Classical quantitative genetics is used to partition the total
variation between replicates into a genetic and a nongenetic component, whereby only the
genetic component is heritable and hence of interest to the breeder. The juxtaposition of
molecular and measurement data—against a controlled genetic background—is the basis of
quantitative trait locus (QTL) analysis, a statistical procedure that detects trait-marker
associations. Molecular markers associated with desirable avocado traits are employed in
marker-assisted selection. Below, we describe two main areas of research that we have been
pursuing over the past 12 months:
(1) Statistical Analyses: We completed the quantitative genetic analyses for our multi-year data
on growth rate, flowering, and fruit load. Breeding advance can only be achieved if the traits of
interest have an underlying heritable (genetic) component. Using Analysis of Variance
(ANOVA), a statistical technique that splits total variation into genetic and nongenetic
components, we determined heritability for three measures of growth rate (tree height, canopy
diameter, and stem girth; data collected 2001–2004), flowering (2003–2005), and fruit load
(2005 only; based on a visual determination of the number of fruit on a tree). Significance for
deviation from the null hypothesis was assessed using F-ratios (from the ANOVA table of Proc
GLM type III sum of squares) implemented in SAS, version 9.1.
(2) Marker Data and Fruit Evaluation: We also devoted considerable time to gathering
additional marker data and linking it to traits measured on our experimental population of trees.
The large fruit harvest this year has provided us with a generous amount of information on fruit-
related traits. Preliminary information on the relationship between markers and fruit traits is
presented; these relationships have not yet been statistically evaluated but serve to illustrate the
principle of marker-assisted selection.
Our statistical analyses concentrated on the assessment of heritability, genotype x environment
interactions, and trait correlations. These statistical measures shed light on the breeding potential
that can be expected when selecting for a given trait. Having used our molecular markers to
determine paternity for each tree genotype, we can also partition the total variation in a trait by
Broad-Sense Heritability: Of all the factors that influence growth rate—as expressed in terms of
tree height, canopy diameter and stem girth—the heritable component accounts for ca. 30%
(Table 2). The higher the heritability value, the faster the trait will respond to breeding. A value
of 30% is respectable, but nonetheless suggests that environmental factors have significant
impact on tree growth. Growth rate averaged 5–6 cm/month, attaining 14 cm/month for ‘Gwen’
progeny genotype 100. Although life history studies generally predict exponential growth in the
early stages of plant development, the growth rates of our trees are linear over the three-year
time interval examined.
Genotype x Environment Interaction: No genotype x location effect was noted for growth rate,
indicating that none of the genotypes shows a marked preference for one site over the other.
Flowering and fruit load showed a relatively weak effect (23.9 and 17.6%, respectively; Table 2).
This means that different genotypes show differential flowering and fruit loads/tree depending on
which location they are growing at (Irvine versus Riverside).
Table 2. Broad-sense heritability and genotype x environment interactions for three measures of
growth rate (tree height, canopy diameter, stem girth), flower abundance, and fruit load per tree.
These values are based on over 90,000 data points.
Tree Canopy Stem Flower
height diameter girth abundance
heritability (%) 34.4 29.7 28.5 32.3 23.4
NS NS NS 23.9 17.6
Trait Correlations: Surprisingly, none of the growth rate measures was correlated with
flowering abundance, and only a moderate correlation was found between growth rate and fruit
load. In practical terms, this means that selection for high fruit yields is not genetically tied to
faster growth. In other words, breeding can focus on combining high fruit yields and short
stature, rather than having to put up with large trees when selecting for high yields. This is a
valuable property, given the trend toward breeding smaller avocado trees. Flowering abundance
was not correlated with fruit load. One might expect higher fruit yields in response to abundant
flowering, but the fact that only one in a thousand fruit attains maturity probably accounts for the
lack of a correlation.
Pollen Donor Effect: Our molecular markers allow determination of the pollen parent of each
experimental tree, enabling growth rates and flowering data to be linked to paternal origin.
Specifically, we can ask whether the type of pollen donor has a measurable effect on selected
growth parameters. Table 3 shows that ‘Gwen’ progeny sired by ‘Fuerte’ was significantly
shorter than progeny sired by ‘Bacon’, ‘Zutano’, or mixed/unknown sources, and produced fewer
flowers. ‘Zutano’-sired progeny had a significantly higher fruit load than the other genotypes,
combined with significantly smaller canopy diameter and stem girth. Moreover, progeny in the
mixed category was significantly taller than ‘Fuerte’- and ‘Zutano’-sired trees and taller
(nonsignificantly) than ‘Bacon’-sired progeny. Similar analyses will be performed on this year’s
fruit evaluation data. Paternity-specific fruit attributes may be identified that would not be
readily detected otherwise.
Table 3. Mean effects of pollen donor on growth rate (tree height, canopy diameter, stem girth;
all in centimeters per month), flower abundance, and fruit load per tree.
Tree Canopy Stem girth Flower
height diameter abundance
‘Bacon’ 5.931(a,b) 6.045(b) 0.226(a) 1.965(a) 1.410(b)
‘Fuerte’ 5.002(c) 6.482(a) 0.213(a) 1.418(c) 1.385(b)
‘Zutano’ 5.774(b) 5.241(c) 0.197(b) 1.846(a) 1.614(a)
Mixed 6.289(a) 6.484(a) 0.223(a) 1.604(b) 1.446(b)
Marker Data and Fruit Evaluation
Six markers were added to our data set between September 2005 and February 2006: AVT386,
AVD003II, AVD006, AVD022, AVD010, AVD028. Since June 2006, we have added
AVD037II, AVD026, and AVD036, taking the total to 16 markers.
By the end of 2005, our marker analyses had revealed that all but one of our 200 tree genotypes
(whose maternal parent is ‘Gwen’) had been outcrossed (i.e., the pollen came from a different
variety) and that 98 of the 200 genotypes had ‘Bacon’, ‘Fuerte’, or ‘Zutano’ as their male parent.
For the remainder, the male parent did not match up with any of the varieties forming part of our
molecular reference archive.
Fruit Weight and Shape
Here we present partial data sets on fruit weight and fruit shape in order to explore trends. Fruit
weights are based on only 1–2 fruits/tree and about 50–60% of trees in respective locations, but
there is no reason to believe that the findings reported below would change substantially once
complete data have been gathered.
For trees at SCREC and AgOps, fruit weight averaged 280 and 226 g, respectively. Table 4
presents a summary of fruit weights partitioned by pollen donor (‘Bacon’, ‘Fuerte’, and
‘Zutano’), suggesting that ‘Bacon’ pollen sires somewhat heavier fruit and ‘Zutano’ pollen
somewhat lighter fruit. At SCREC, ‘Fuerte’-pollinated progeny produced smaller fruit than
‘Bacon’, but the reverse was true at AgOps, a possible consequence of the smaller sample size at
the latter location. The larger ‘Bacon’-sired fruits at SCREC could also signal better adaptation
of ‘Bacon’ progeny to the coastal Orange County conditions than to the Riverside climate, or
may be related to the larger size of ‘Bacon’ trees reported above. A comparison of the average
fruit weights for these three pollen sources versus the overall orchard averages, suggests that at
SCREC the genotypes in the “other”-category are likely to be larger-fruited than the fruit sired
by ‘Bacon’, ‘Fuerte’ or ‘Zutano’, whereas the reverse is true at AgOps.
Table 4. Average fruit weights [grams], with sample numbers in parentheses.
SCREC AgOps Both locations
‘Bacon’ 280.6 (48) 257.8 (41) 269.2 (89)
‘Fuerte’ 268.3 (46) 264.4 (27) 266.4 (73)
‘Zutano’ 266.8 (38) 248.2 (30) 257.5 (68)
Overall 271.9 (132) 256.8 (98) 264.4 (230)
Fruit shape was scored using the IPGRI descriptors and partitioned into four main shape
categories representing spheroid or somewhat spheroid (“2”), obovate (“6”), narrowly obovate
(“5”), and somewhat clavate (= elongated; “8”). Table 5 illustrates the relationship between
pollen source and fruit shape score. A majority of genotypes are in the intermediate shape
categories “5” and “6” that characterize ‘Gwen’ fruits and would be expected in ‘Gwen’ progeny
trees. The more extreme round or elongate shapes arise much less frequently. Upon closer
inspection, these data reveal a slight tendency for ‘Bacon’-sired progeny to produce rounded fruit
(mostly shape categories “2” and “6”) and ‘Zutano’-sired progeny to produce elongate fruit
(mostly categories “5” and “8”).
Table 5. Fruit shape, scored using IPGRI descriptors. Values are counts for AgOps and SCREC
“2” “6” “5” “8” Totals
‘Bacon’ 31 27 21 10 89
‘Fuerte’ 26 11 20 15 72
‘Zutano’ 10 9 30 18 67
Total 67 47 71 43 228
The Role of Alleles
The data presented so far are interpreted in terms of paternal influence acting in a ‘Gwen’
maternal genetic background. We can focus on one more level and examine relationships
between traits and particular alleles. Because a given tree genotype has two alleles at each
genetic locus—one allele from each parent—and because each parent in turn has two alleles
available to pass down to its offspring (e.g., A or B from the mother and C and D from the
father), the same parental combination can result in four different progeny genotypes (AC, AD,
BC, or BD). Table 6 illustrates a concrete example of fruit weight partitioned by alleles present
at microsatellite marker locus AUCR418.
Table 6. Relationship between fruit weight and allelic composition for microsatellite locus
AUCR418. Alleles “e” and “h” are present in ‘Gwen’ and ‘Fuerte’, alleles “h” and “c” in
‘Zutano’, and alleles “d” and “g” in ‘Bacon’. All fruit weights [grams] are averaged for all trees
(n = 111) possessing the allele in question.
Allele “e” Allele “h” Average
Allele “e” 334.42 271.53 302.98
Allele “h” 271.53 288.16 279.85
Allele “c” 249.69 240.00 244.85
Allele “d” 249.85 294.53 272.19
Allele “g” 274.20 291.91 283.06
Average 275.94 277.23 276.58
The marker locus itself is not (usually) synonymous with the gene affecting fruit weight.
However, its proximity on the chromosome to a gene that does control fruit weight determines
whether the marker can detect any signal. In the example illustrated in Table 6, genotypes
having the allele pair e/e (two identical copies of allele “e”) produce the heaviest fruit (334.42 g),
whereas genotypes with allele pair c/h have the smallest fruit (240.00 g). Alleles “d” and “g”
(from ‘Bacon’) produce larger fruit when combined with ‘Gwen’ allele “h” than in combination
with ‘Gwen’ allele “e”. If this trend were to prove statistically significant, then marker-assisted
selection for large fruit size would recommend retention of seedling genotypes having the “e/e”
genetic constitution but removal of seedlings with the “c/h” constitution.
This example serves to demonstrate the tremendous power of genetic markers in peeling away
the oft-confusing veneer of phenotype to reveal the genetic underpinnings of any given trait for
which sufficient data have been collected. The more genetic markers we can add to our
database, the greater the likelihood of detecting a signal from genes affecting a trait of choice.
The more numerous and diverse the traits measured on the trees, the more can be learned about
the genetic structure of the avocado genome. Both our microsatellite markers and our trees
represent a substantial resource that should be tapped. The permanence of the marker data
means that, once completed, they will be available for referencing all future traits measured.
Conclusions and Timeline
The labor-intensive nature of this project is clear. Table 7 summarizes the number of data points
that factored into the non-molecular matrix alone. In the coming year, we hope to boost the
number of markers in our molecular data set.
Table 7. Updated table on labor requirements for collection of data on growth rates, flowering,
and fruit evaluations. Readings = number of measurements taken as part of the fruit evaluations.
Noninvasive evaluations include fruit shape, weight, width, length, horizontal circumference,
vertical circumference, and skin texture. Invasive measurements include ripe fruit weight, 4 seed
attributes, 4 skin attributes, and 4 flesh attributes.
Evaluation type # Years Readings # Fruits per Tree # Data points
or year per trait tree
Tree height 5 1 n/a 700 3,500
Canopy diameter 4 2 n/a 700 7,000
Trunk diameter 4 2 n/a 700 5,600
Flowering 3 6 n/a 700 75,600
Fruita 2005 7 1–15 62 6,510
Fruit 2005 13 1–3 62 2,418
Fruit 2006 7 2–8 525 29,400
Fruitb 2006 13 1–3 525 20,475
Hours/evaluation: Noninvasive: ca. 50 fruits in 3 hours = 16.7 fruits/h [3.6 min/fruit]
Invasive: ca. 30 fruits in 3 hours = 10 fruits/h [6 min/fruit]
Accordingly, we have drawn up a tentative timeline for data collection and subsequent analyses.
While tree growth and fruit data are adequate for preliminary studies (heritability and genotype x
environment interaction effects), this is not true for the molecular data: a QTL analysis is not
normally initiated until allele data are available for a threshold number of markers. This
threshold is a function of the distribution of QTLs across the chromosomes and of an organism’s
chromosome number: the greater the number of chromosomes (avocado = 12), the greater the
number of markers needed. The processing of 40 genetic marker loci over the next 12 months
will enable a first pass-QTL analysis, plus additional time required for data formatting and
manipulation. This translates into a time frame of ca. December 2007. The greater the number
of markers processed, the greater the likelihood of detecting a signal. Our goal is eventually to
run all 127 markers on all experimental trees.
Marker-assisted selection can start as soon as QTLs have been found. QTLs are visualized in
terms of the associated alleles (“bands on a gel”) and are ranked by efficacy, which assists the
plant breeder’s decision which QTLs to select on. Ranking depends on several factors, including
strength of the marker-trait association, correlation between markers, and overall allelic
composition of the breeding material. As a long-term strategy, we recommend that marker-
assisted selection be applied to our Gwen progeny trees for two successive generations.